DSP: The Discrete Fourier Series of Periodic Sequences

Digital Signal Processing

The Discrete Fourier Series of Periodic Sequences

D. Richard Brown III

D. Richard Brown III 1/9

 DSP: The Discrete Fourier Series of Periodic Sequences

Big Picture

finite-length periodic
sequence sequence
periodic extension
x[n] x̃[n]

DFT DFS

DTFT X[k] X̃[k] DTFT

X(ejω ) X̃(ejω )

D. Richard Brown III 2/9

 DSP: The Discrete Fourier Series of Periodic Sequences

Review: Fourier Series of Continuous-Time Periodic Signals

Suppose we have a periodic continuous-time signal x̃(t) with period T0
2π
such that x̃(t + rT0 ) = x̃(t) for all t and all integer r. We denote Ω0 = T0
as the radian frequency corresponding to the period T0 .

Under certain conditions satisfied for most signals of interest in signal

processing, such a periodic signal can be expressed as a sum of complex
exponentials with frequency 0, Ω0 , 2Ω0 , 3Ω0 , . . . , i.e.,
∞
1 X
x̃(t) = C[k]ejΩ0 kt
T0
k=−∞

where C[k] are the (usually complex) Fourier series coefficients. These
coefficients can be easily computed by observing that the set of functions
{ejΩ0 kt } for k ∈ Z is an orthogonal basis, leading to
Z
C[k] = x̃(t)e−jΩ0 kt dt
T0

D. Richard Brown III 3/9

 DSP: The Discrete Fourier Series of Periodic Sequences

Basis Functions for the Discrete Fourier Series (DFS)

The principle of the discrete Fourier series (DFS) is very similar to the
continuous-time case. We assume a periodic discrete-time sequence x̃[n]
with period N samples.

We denote ω0 = 2πN as the normalized frequency corresponding to the

period N and consider the set of discrete-time periodic complex
exponential sequences

{ejω0 kn } = {ej2πkn/N } = {ek [n]}

with k ∈ Z and n ∈ Z.
D. Richard Brown III 4/9
 DSP: The Discrete Fourier Series of Periodic Sequences

The DFS is a Finite Sum

Observe that, for integer r, we have

ek+rN [n] = ej2π(k+rN )n/N = ej2πkn/N ej2πrnN/N = ej2πkn/N = ek [n]

hence our original set {ek [n]} for k ∈ Z has many redundant elements. In fact,
there are only N distinct periodic complex exponentials in the set, which we can
choose as
{e0 [n], e1 [n], . . . , eN −1 [n]}
for n ∈ Z.
Along the same lines as the continuous-time Fourier series, we can then write the
DFS
N −1 N −1
1 X 1 X
x̃[n] = X̃[k]ek [n] = X̃[k]ej2πkn/N
N N
k=0 k=0

where X̃[k] are the (usually complex) DFS coefficients. A key difference with
respect to the continuous-time Fourier series is that the DFS of a periodic
sequence x̃[n] can be written as a finite sum.
D. Richard Brown III 5/9
 DSP: The Discrete Fourier Series of Periodic Sequences

Computing the DFS Coefficients X̃[k]

We have
N −1
1 X
x̃[n] = X̃[r]ek [r]
N r=0
If we multiply both sides by e∗k [n] and sum from n = 0, . . . , N − 1 we can write
−1 −1 N −1 −1 N −1
N N N
!
X
∗
X 1 X ∗
X 1 X ∗
x̃[n]ek [n] = X̃[r]er [n]ek [n] = X̃[r] er [n]ek [n] .
n=0 n=0
N r=0 r=0
N n=0

Note that
N −1 N −1
(
1 X 1 X j2π(r−k)n 1 r − k = mN with m an integer
er [n]e∗k [n] = e =
N n=0 N n=0 0 otherwise.
Hence
N
X −1
x̃[n]e∗k [n] = X̃[k]
n=0

which is similar to the CT case with an integral is over one period of x̃(t)e−jΩ0 kt .
D. Richard Brown III 6/9
 DSP: The Discrete Fourier Series of Periodic Sequences

Some Properties of the DFS

D. Richard Brown III 7/9

 DSP: The Discrete Fourier Series of Periodic Sequences

Periodic Convolution Property of the DFS

Recall, if we had signals xi [n] with the necessary DTFTs we could write
∞
X
x3 [n] = x1 [k]x2 [n − k] ⇔ X3 (ejω ) = X1 (ejω )X2 (ejω ).
k=−∞

For the DFS, we have have this relationship only for periodic convolution:
N
X −1
x̃3 [n] = x̃1 [k]x̃2 [n − k] ⇔ X̃3 [k] = X̃1 [k]X̃2 [k].
k=0

Remarks:
1. If x̃1 [n] and x̃2 [n] are both periodic with period N , then so is x̃3 [n].
2. Periodic convolution specifies a sum over one period.
3. Since all of the signal are periodic, x̃2 [n − k] = x̃2 [n − k + mN ] for
integer M . For example x̃2 [1 − 3] = x̃2 [−2] = x̃2 [N − 2].
D. Richard Brown III 8/9
 DSP: The Discrete Fourier Series of Periodic Sequences

Periodic Convolution Example

Suppose the periodic sequences x̃1 [n] and x̃2 [n] are generated as periodic
extensions of
x1 [n] = {1, 2, 3}
x2 [n] = {a, b, c}
with N = 3. Then the periodic sequence x̃3 [n] can be computed as
2
X
x̃3 [0] = x̃1 [k]x̃2 [0 − k] = 1 · a + 2 · c + 3 · b
k=0
2
X
x̃3 [1] = x̃1 [k]x̃2 [1 − k] = 1 · b + 2 · a + 3 · c
k=0
X2
x̃3 [2] = x̃1 [k]x̃2 [2 − k] = 1 · c + 2 · b + 3 · a
k=0

and all other elements of x̃3 [n] are implied through the periodicity.
D. Richard Brown III 9/9